OFFICIAL

>> ALISTAIR: Okay so last year we did a session for digital leaders week and we framed it as a bit of a workshop
0:13
conversation around how we build development communities and collaborate on things and it went
0:20
down quite well with a few people who turned up but one of the bits of feedback we got was we should have more examples in the mix
0:26
to help start conversations so this year we are focusing on data modelling and children's
0:32
services um it's a topic that keeps coming up in all the different kind of meetings and
0:38
groups that I participate in and it's something that we're running a project on at the moment and
0:44
what we have done is pull together a few kind of exemplar projects from people who've kindly agreed to
0:49
speak this afternoon we'll do a few presentations and hopefully leave a bit of time at the end
0:55
for either questions directly to individual presenters or discussion or people to just raise
1:00
problems or thoughts or comments or whatever else they might want to speak about on this subject and then obviously we'll
1:07
hopefully advertise at the end our spaces for local authority colleagues to
1:13
continue those conversations into the future so agenda for today I’m just going to
1:20
introduce things as I am doing now and talk for a few minutes about data to insights
1:26
journey with data modelling so far and then pass over to some presentations from colleagues in
1:33
dudley um that's Inderjit Lahel, East Sussex Alaister lee and Surrey um whose 
1:41
council name i've missed off the list there sorry, Sara Dicerto, and they'll each give an a kind of five
1:47
ten minute presentation on something interesting that's been going on in their local authority and that
1:52
might be useful to think about for other of the councils and then hopefully a bit of time at the end for a
1:58
bit of discussion so without further ado i passed off into the um
2:03
the data to insight bit it's about modelling placement demand, and the story starts really with a
2:10
completely different problem and one of the first things that data to insight did
2:15
once it got um got some kind of full-time resource established last year was start a project to model likely
2:22
demand for front door services and social care at the end of 2020's lockdown period
2:27
and we did a project with local authorities and produced a tool which lots of people found very useful
2:32
um some authorities found it gave them quite accurate or timely predictions
2:38
about what might happen in their front doors when lockdowns ended um and we then tweaked that and refined
2:45
it a little bit for earlier this year the first lockdown of 2021 and again it was quite useful but the
2:50
longer this period of time goes on the less long-term usefulness this kind of tool has
2:56
partly to do with the kind of the way we were looking at historical data and defining what counted as normal and
3:01
what counts as a deviation from normal and what have you and partly just for other reasons but it
3:08
clearly demonstrated there was a lot of appetite for this kind of work um hence the workshop today to talk
3:14
about it more so we're currently starting a follow-on project which is a lot more
3:20
focused and the way we got there was we took the original piece of work and went back to get feedback
3:26
and then went out to do some discovery work with local authorities to ask about what kinds of modelling or
3:33
predictions or forecasting or just understanding their data did they want to be able to do what they
3:39
couldn't really do well locally or do they think was really important or most important of all did they think
3:44
would lead to an actionable decision so they would do something differently having seen the model data
3:52
and we brought lots of ideas out of that but the key point was that we wanted to do something specific
3:58
and kind of self-contained that would demonstrate how useful this can be in relation to a kind of a decision you
4:05
can make differently having seen the data something you'll do based on that information and modelling placement demand came out of
4:12
that user research as a consistent need across every LA we talked to
4:18
so our key brief at the moment and we're in the first bit of design and build at the
4:23
moment having completed that user research is about forecasting demand for placement categories
4:29
so that LA’s can ensure there's supply to meet demand and they're getting best value and what we want to do with that is um
4:36
use historical data sets and time and kind of up to the minute data to supplement those
4:42
and build out a placement demand forecast from that based around understanding
4:49
what the useful segmentations of that data are so we're not just saying you need 500 placements in your local
4:54
authority or you're going to need 520 we're saying which kinds of placements or which kinds of children or which
5:00
kinds of situations lead to which kinds of demand and that's the place we're at the moment
5:07
so the approach is to build the very simplest thing we possibly can that will model some demand and will work based on that historic data
5:14
and we can test it using historical data as well to see how well it performs in that area easy
5:19
for local authorities to put data into because data to insights purpose is to build tools that people
5:24
can use easily and save some time rather than creates work for them and helps them see which segments are
5:30
the drivers of future demand and then take that and iterate it over the remainder of the project to
5:36
see where we can take it I’m kind of flying through these slides but I can make them available
5:41
afterwards if people want to read all the detail so this is our roadmap and the important
5:47
thing to understand about this is we've got a very small amount of stuff on the left of the screen saying the minimum viable
5:52
product we want to produce takes data from the ssda903 models placement demand from new starts and
5:58
from placement transitions and allows a user to then model future demand from that
6:05
the middle bit says when we release this to our kind of our users as a first public release then we
6:13
will be doing that plus something else from the right hand side of the screen and the right hand side of the screen is
6:19
all of the ideas of things we think we'd be able to then expand into once we can do that first thing and so
6:25
all of those things are things we'd like to do we'll figure out which thing is the best fit for our users to add in with the
6:30
time we've got to build a tool as I say self-contained and really just useful in making a decision locally
6:37
the last thing I wanted to speak about on this is how we're delivering it and the data to insight model really is about collaborative development and
6:44
about pooling resources across local authorities so where there are things that we all need to do or lots of us need to do
6:51
in similar ways and we can brook a little bit of compromise about the details of how it's done then we can
6:56
tap into economies of scale we can tap into funding streams that we wouldn't be able to get into as individual local authorities
7:02
and we can develop things that we wouldn't be developing otherwise and the kind of development model we're
7:07
trying to prove here is the same one we used in the first phase of the project where we have a very small core resource
7:13
and we're working with a partner called Social Finance a non-profit technical team
7:20
small resource from data to insight and then there's a kind of a small circle of local authorities around that who are
7:26
involved from start to finish and then a slightly bigger circle of local authorities who dip in at
7:33
different points to give us fresh eyes perspective on the project as we go and make sure we're not going down an unuseful rabbit hole
7:39
but then there's a whole wider community around the outside who are welcome at all sages to kind of keep tabs on
7:45
what we're doing feed in their ideas and their thoughts and of course use the product for free once it's finished and keep sending that
7:51
feedback and keep helping to iterate it into the future the key thing to say about building in this way is more
7:57
difficult than just building at all but the potential benefits are bigger as well because we don't just
8:02
gain we don't just benefit from skills that exist we bring those skills into our community and we're able to use them
8:09
and reapply them to different projects and different purposes once we've got them
8:15
um so one of the key partners on that sorry is um we've got Birmingham council, Sutton
8:20
council and Dudley are our regular feedback circle for this first bit of this project and that brings me nicely onto Dudley
8:28
council um and we've got Inderjit Lahel here to do the first of our um overviews of a kind of a case study
8:36
of something that's happening in a council at the moment that might be interesting for people so Inderjit if you're there I will hand over to you 
>> INDERJIT: yeah, should I um I would in slides should i just upload
8:48
>> ALISTAIR: yeah, if I let you share your screen then It’ll be easier for you to drive them for me to drive 
>> INDERJIT: cool
9:01
>> INDEJIT: can you see that?
>> ALISTAIR: yeah yeah 
>> INDERJIT: cool 
>> ALISTAIR: thanks 
>> INDERJIT: cool okay
9:09
Hi everyone my name is indeed Inderjit Lahel, the head of integrated commissioning performance and partnerships for Dudley council
9:14
so I have keen interest in this around i guess one being head of commissioning
9:20
for adults and children's um social care including a CND and but also the lead for performance
9:26
and intelligence as well within the council so it's it kind of hits a fair few of my
9:32
areas really so what we what we what we tried to do in Dudley um
9:39
uh was was look at what you know what uh what we've got in the system so a bit of background really for us so we've got
9:45
um a number of looked after children… It’s fairly consistent it's you know it's
9:52
it's between like you know 620 to upwards of the high sixes
9:57
and spending within residential placement is growing as i guess it isn't a lot of authorities
10:03
and we've got some good work going on around supported accommodation so we're looking at developing a new model
10:08
and around a partnership model with the voluntary steps that's quite exciting um we've got a
10:15
number of internal residential homes in Dudley and then you've got the whole kind of um the fostering piece where we've got
10:21
internal fostering we've got external fostering as well um and you know what are they I guess
10:26
you know as a commissioner it's like what is the true cost of that and what's the best what's the best way forward in terms of our fostering offer so just a little
10:34
a little bit of background to where we are in Dudley, um so what we did so one of my analysts
10:41
and one of my commissioners took on an exercise so we looked at where we are in Dudley terms of numbers
10:46
our spend across each of the areas in our growth um over the last five years our aim was
10:53
to develop a tool um so we could start look at things a little bit differently
10:58
um so around looking at placement types across our population um the aim was to look to
11:06
change if we could if we could project and change different factors within our tool
11:12
um to change the i guess the um the journey of children using a different approach so my thought
11:18
process around this was that if you know if social care adopted a fundamental change model
11:24
operationally you know what would that mean on the journey of a child and our interventional models that we
11:30
need to put in. Ultimately we'd look to um influence our spend
11:35
across our population and then the key bit from a commissioning perspective was to start to have
11:40
different very different conversations with that with our market i think you know at the moment and as it
11:45
is nationally the market is provider-led um and what we want to
11:51
do is like start to use our models and our tools to start to change conversations have them more strategic conversations
11:57
with some of our different providers so we started that a little bit with a couple of providers
12:02
it's kind of stalled a little bit and i'm hoping that this piece of work gets it kind of back on as well
12:08
So where did we get to so we developed we developed a raw tool um still kind of needed a lot of work to
12:14
it we shared that with them with Alistair and and Michael from Social Finance
12:20
um we got you know we looked at interventions how we could implement these
12:25
but we didn't really get to progress them well but what it has given us it's given us a bit of a baseline to start to
12:30
kind of compare where we are across that compared to LA’s so we looked at you know what our budget is for um
12:36
placement spend across different areas and whether like we know we're comparable to that our internal versus
12:41
external and some of that costs um so in terms of how it links into the
12:48
data to insight work um so it's really hard to do this because I’m not showing you the tool
12:53
but um it's just a bit of an overview so we see this piece of work that that Alistair and
12:59
Michael from Social Finance is doing is that's really key really so it for me it's that it should start
13:06
to shape our commissioning intentions um i think we can start to look at where
13:11
we spend what the outcomes are for internal and external and i think as i said before the interventions and influence over
13:17
social care model is quite key so i think you know from a commissioning background it's
13:22
you can put you can put the best commissioning models out there that you that you that you wish for really
13:27
but if you haven't got the um the matching operational model or the matching operational kind of protocols
13:34
etc it's it doesn't work it has to be part of a more strategic approach i think what Alistair was saying before around
13:40
you know the model the work that they're doing around influencing and making
13:46
decisions i think that's quite key for me and we think that's this can help us as well around us efficiency strategies.
13:52
we've got our efficiency strategy which is currently being updated um but i see this as being quite key to
13:58
developing consistent sufficiency strategies across the country really and as i said before um talking to
14:04
providers around different models um and how that can work in in different areas and looking
14:10
towards outcomes based and commissioning if possible. I think as a last point i think it should help us to
14:16
pull away from some of our um scattering approaches to purchasing when i say purchasing um
14:22
kind of purposefully really, because it's purchasing rather than commissioning and so I’m hoping that the work that
14:29
we're doing here should start to help us have a different a different conversation a different commissioning
14:34
model um i think bear in mind i come from a predominantly an adult's background and you know the commissioning is very
14:40
different in adults and children's and so I think i'd say that the adults one's a bit more kind of
14:46
um i guess a bit more kind of developed and i think um but there's massive opportunities in
14:51
children to do something differently so that's my bit Alistair i know it's not kind of
14:57
mad data in there but i think it gives hopefully people a bit of an overview of what we tried to do and it
15:03
did you know it hasn't progressed as much as we wanted it to um but i'm hoping that the work we're doing
15:08
with data to insight will really kind of push it forward 
ALISTAIR: i think that's really useful thanks Indy
15:15
um shall we pass across to um Alastair now
15:21
um Alastair lee, we have two Alistair’s involved in data to insight um
15:26
and Alastair Lee is the first of those two he's going to talk to us about East Sussex's work on SEND
15:32
>> ALASTAIR Lee: um can you see that alastair you got it 
>> ALISTAIR: yes 
>> ALASTAIR Lee: great stuff. good afternoon everybody so I’m Alastair Lee I'm children's services data and
15:38
information manager for East Sussex county council and actually work quite closely, very closely with Alistair
15:44
on the data insight project um so but this this is something that predates data insight by some years
15:50
um so uh
15:56
there we go so the problem um we identified in 2014 um where
16:02
the number of of EHC plans (Education Health and Care Plans) in East Sussex was increasing significantly and that was
16:08
uh partly due to the extension to 25, up to 25 but not entirely there was other things going on
16:14
that we wanted to know um and it was a real sense of you know this is kind of getting out of control we don't understand it
16:20
we need to get a grip of it and and look at the long-term impact what's going on so myself and my team with our data
16:28
hats on um talk you know could we take historical SEN data and demographic projections and
16:34
produce something of use to the service and managers and we produced the first set of data um in 2015
16:40
um and we've been running ever since um and it's proven quite successful and i'll just talk you through um
16:47
the how we've done it so the data is fairly straightforward and something that you probably have already if you
16:52
work in children's services so it's the school census number of children with statements EHC plans broken down by
16:58
uh abandoning severity that we have locally so A to E which is a local banding severity
17:04
age um an age allows us to do both age but also school age and a categorization of the
17:10
11 primary need types so interesting we're not using send two we're using um school census here but that's that's
17:16
the data we use um then we have the population projections um to forecast the future population and
17:22
that's the starting set of data um and the basic assumption is that uh
17:29
things should stay things will stay the same then actually as um yeah the total assumptions that current
17:35
prevalence rates for different primary need types will remain unaltered so as uh you have x percent of the
17:42
population has got ADHD and as the population increases that number increases and that that kind of
17:47
is is the basis of the the initial assumption um but that's not true over over the periods with data
17:55
we've got we can see that some are increasing more than others and some are dropping off more than others and so that
18:00
is factored into the model as well so we have a five-year trend of what the the trend is either upwards or
18:05
downwards or fairly flat um and so all those those three things
18:10
are factored in so we've got the school census data we've got population projections and then we've got five-year trend factors it's all
18:17
pulled in that's then stuck on some charts so we do the maths and we make some
18:22
pretty charts and this is a the most recent overall um version for the different uh prima need groups
18:30
the yellow stuff is historical so that's you can see back to 2014-15 where we started that's all now actual
18:37
data but then forward you've got the real data and as you can see
18:42
um ASD (Autism Spectrum Disorder) is going up and then equally you've got um SEMH is
18:49
going up so that was a bit of a surprise to everybody in terms of how big those those changes were
18:56
What we then built into the model uh and it's not just a mathematical model is that we needed
19:02
service refinement we needed a challenge from the service to say this you know some of those lines are stupid
19:07
don't be ridiculous go back and think again and so we ran initially a workshop it
19:14
was a day-long workshop and went through every indicator um at every age group um and got them to
19:20
say yes that looks right, it's a bit steep, it's a bit low and we refined the
19:25
the measure so you can see in the middle of this um the picture in the bottom you've got the
19:31
agreed trend factor so that is the one agreed with the workers it's not one that we've calculated
19:37
you can see above that the average yearly change is 2.7 but they've actually agreed a slightly higher one
19:44
uh to do that and then uh beneath that you see the various charts and what we do in the meeting is we've got
19:49
that's dynamic so um uh Joe my colleague can actually change the numbers and get them to look at what the
19:56
trend does um with that then uh has a really good discussion with the work the staff about what
20:02
what this means for them and does it feel right to them as professionals um once that's all agreed that then
20:09
forms the basis of the forecasts. Each year we do an annual accuracy check
20:15
so we've been doing that ever since 2014, 2015 and generally speaking we are
20:20
either it's between plus or minus two point six percent out this year we were under forecast by
20:27
point two percent uh so that's the actual total was three thousand six hundred thirty three compared to two thousand six hundred
20:32
twenty four and that spread between the age groups and the primary needs um for example ASD was under forecast by
20:40
5.9 percent that was 66 pupils so in 2019/20 there were 1053
20:46
we predicted 1017 but actually the actual was 1183. um and that
20:52
threat that under forecast is part of that and the workers and we had set the trend so that's where we got that data from
20:59
It's not a perfect science um and it's the trends over time that are as important as the detailed actual
21:04
numbers in helping the service planning so is it going up how steeply is it going up and if it is going up that steeply
21:10
what does that mean longer term um also it enables you to think about what
21:17
interventions you could put in so if it's going to go up is there an intervention we could put in that would bring it down and you could
21:22
then see the impact of some of that hopefully um so how is it being used does it has had
21:29
an effect do we continue to use it so we keep producing it. We have run the workshops every year the
21:36
workshops have got shorter as we've got used to it but the real impact of this is that in a couple of years we're going
21:44
to open two new special free schools um to deal with ASD in east sussex, one in Eastbourne and one in hastings
21:50
um and the forecasting data is as my colleague the place planning manager
21:55
we continue to use the forecast to turn the next stage of our strategy to create more percent provision in east Sussex
22:00
so quite a simple mathematical tool but it's a mixture of that and actually working frontline staff to
22:06
understand um to get their sense of what is changing let's build the model that then um
22:12
enabled us to do this and do this work so that's a very quick run through um i think that's all from me
22:21
>> ALISTAIR: that was a marvellously speedy run through thank you 
>> ALASTAIR LEE: As ever Alistair
>> ALISTAIR: [laugh] um so a little bit earlier than planned
22:28
then um can i pass over to sara from surrey who's going to talk to me about her um
22:34
pathway modelling and before i do so sorry I’ll just welcome new people who've arrived in the meeting
22:40
since we started. Um we're just running through a few presentations and kind of overviews of
22:45
examples of data modelling practice in different councils at the moment and then we'll have a little bit of
22:51
space for questions and comments at the end. Um we're recording the session so if you
22:56
don't want to be recorded then now is your opportunity not to speak but if you're quite happy then otherwise
23:02
feel free to pop a message in the chat along the side or pop your hand up when we get to the questions section
23:07
Later. Um Sara are you there to take over from me.
>>SARA: yeah um I will just share my screen
23:24
has my screen loaded correctly 
ALISTAIR: yes it has can see that thank you 
SARA: wonderful thank you so much um so good
23:31
afternoon everyone um i am Sara from Surrey county council and i am here to present about
23:37
a model that we created about a year ago um for our local authority so uh
23:44
about a year ago it was effectively um well not the beginning of the pandemic but a few months in and Surrey county
23:50
council like the vast majority of local authorities had a very important need to understand the
23:56
demand and the capacity for social care services that would be
24:01
projected into the following few months to try and understand whether there were any criticalities
24:08
that we needed to address and whether our staffing arrangement for specific tasks to be carried out
24:14
within our local authority was adequate for the projected need
24:20
we needed to do some on the basis of the of different scenarios um at the time obviously we had no real
24:27
understanding of whether there would be one two or multiple lockdowns and we needed to account for the possibility
24:33
that there might be a few there might be none they might come at a different time
24:38
um so in order for us to cope with the need to basically understand our projected demand and our
24:44
projected capacity, we built a model that related the demand that we were projecting with the
24:50
capacity that was projected to be available for different teams um and that was crucial for the council
24:55
to make very strategic decisions about using resources effectively and shifting people across them
25:01
when there was an identified need. Um so the way we went about building the
25:07
model was effectively using data sets from different teams um i will run you through the structure
25:12
of what we did very quickly um so basically Surrey county council has an offer of social care that
25:18
goes from uh level one to level four um as a windscreen of uh support basically
25:25
um and we start with our children's single point of access uh that is available
25:30
universally um the uh level two uh aspect of our uh offer is basically
25:37
our early help offer both internal and external um and level three is the more um
25:43
social care um in a traditional sense as it were um so children need child protection
25:48
plans uh and children with disabilities um and level four is around care
25:53
Proceedings, children who are looked after and care leavers so in essence basically um the general
26:00
understanding the general sort of structure of the model um was that we knew that effectively uh
26:06
the demand that comes through our children's single point of access then it trickles down either into early help or social care
26:13
and a proportion of our social care contacts that then become assessment um mean that a child becomes a child in
26:21
need or is subject to a child protection plan or is all becomes basically looked after
26:27
as a result of care proceeding um and uh the a proportion of those
26:32
who become looked after um either is reunified or ages out of care and becomes a
26:38
caregiver so we took the data sources for all the different
26:44
teams and we looked at how indicators related to one another and built a model that effectively
26:50
we had a number of uh delays and uh trickling down a proportion from each
26:56
level into the following level um gave us an understanding of how uh contacts
27:01
in the children's single point of access related to the number of new child
27:06
protection plans for example or to the number of children who became looked after. As i was mentioning at the
27:12
beginning of the presentation we needed to look into the possibility of different scenarios in terms of different
27:17
lockdowns and um if and when this might happen so we created a long list of 12
27:24
scenarios um and our leadership selected three for further development that were deemed to be the most likely
27:30
scenarios that we could face um and we then developed uh the models based on an
27:36
assumption of uh the influence that lockdown would have had on our demand based on the change that we had observed
27:44
in the previous uh three months when the model started to be developed so basically we took
27:49
April, May and June data that um where from our first lockdown and looked at the
27:55
difference uh with previous months and we the same months in the previous year to understand what the likely impact
28:01
of the lockdown itself had been on the demand for different services
28:07
let me talk you through uh scenario one so if i go on to the first tab
28:14
and i zoom out you will see that the colours that you have looked at in the first structure as it
28:21
were replicated in our overall structure of the different indicators starting from level one access and
28:29
ending with level four support um for each of those indicators um
28:35
let me pick one of basically oops
28:41
oh geez we have a line of actual data
28:48
um and this is how our historical data on which basically we built our baseline projection
28:53
um and then we created a darker blue line that was our COVID19 projection
28:58
and we then kept updating uh the data as he came through um to uh look at how closely our projection
29:06
matched the reality of uh contacts to social care in this particular case
29:12
you can see that on top of the historical trend line plus the projection there are another
29:17
green, amber and red line that are effectively the
29:22
capacity for the team to deal with contacts um i have included in this model um the
29:29
possibility for people to modify these lines by modifying the indicators they are based
29:34
on. So the lines that you are observing now are an example and they are based on a maximum 15 contacts per day
29:41
dealt with by people occupying four full-time equivalent positions
29:47
assuming that um I wanted to change this number and say that actually the contacts they can deal with per day
29:53
are 20. This automatically changes my projection to um show me whether the demand is likely to
30:01
meet the capacity um or there are going to be any issues with facing uh the demand that appears in the system
30:10
you can do obviously the same in terms of the number of full-time equivalents which allows you to make
30:16
hypotheses in terms of um if there were an extra two members of staff what would that mean in terms of in
30:23
terms of the capacity and uh the demand and would that give us what is required in order to face the
30:30
demand that we are likely projecting as possible and likely in the system.
30:36
Um each indicator as i was saying is based on the indicators that precede it basically
30:42
in the tree, as it were, um we adequate stops and uh breaks in terms of the time it takes
30:49
for the different um stops um to move into the following step in the system
30:56
at the very top you can see that there is a description of what the scenario looks like and what
31:02
it is based on in terms of when restrictions will be lifted and if there are any new lockdowns um
31:09
included in this particular scenario here um and also a rating of the likelihood
31:15
of this particular scenario materializing based on the assessment of our
31:21
leadership team what the model basically has allowed us
31:27
to do is shift capacity between teams um when there was an observation that a certain
31:33
team maybe had a little bit extra capacity compared to the demands that they were experiencing uh
31:38
they could be moved across team in order for them to support other teams that
31:44
were struggling and that at that specific point in time um and it became apparent rather quickly that
31:49
beyond managing the pandemic itself the tool would have become useful as a general planning tool
31:55
for staffing and modelling of demand so as you can see from now um
32:01
basically the uh model was born as an excel based spreadsheet but we are currently in the process of
32:08
migrating it to a tableau based solution so that it will be possible uh for the
32:13
model not to have to be updated manually every time um and the assumptions of the model will
32:19
basically adapt themselves based on the new data that is received automatically as the model will start
32:26
feeding off of live data sources so that's our expectation and we should be able to have the new
32:33
version of the model that is based on live data probably around the end of summer now.
32:48
>> ALISTAIR: that's brilliant thank you very much Sara um and thank you especially because um
32:54
one of the things that i think came out of the piece of work we've just done in Data to Insight around data modelling and demand modelling
33:00
and figuring out what should we do with our next bit of this project how do we take it forward and make it useful to local authorities
33:07
is this idea that there's different reasons that you might model there's different kinds of data modelling
33:12
you might do there's different um products that might come out of it and often
33:18
when you say data modelling to someone they'll think oh he's talking about forecasting or oh she means doing scenario modelling or
33:25
or what have you could be doing any one of these things there's a big value i think in
33:30
just in mapping system flows and understanding how the system works so that you can then play that back to
33:35
your leaders and let them make more informed decisions and have a better sense of when things change within that system
33:41
but equally things like that scenario modelling um it's increasingly valuable when we in
33:48
periods of uncertainty. so yeah thanks very much for that um that is the end of our showcase of
33:54
presentations and we've now got about 15-20 minutes to take questions um
34:00
have a bit of a conversation and take input from anyone else who's got anything they want to feed in at this
34:05
point. I’ll kick off with a question from the conversation
34:11
pane there from Joanne Harris which is a detailed question about the SEND
34:16
modelling for you Alastair Lee asking about if you only use the score census data
34:22
and if this is because you don't have a lot of out of burrough replacements or that the outer bouroughs isn't a big 
34:28
factor for you.
>> ALASTAIR LEE: Um and the answer to that, i was typing as you asked me the question right okay
34:33
we have relatively few out of boroughs because we're large so at East Sussex county council we border um Kent, um Surrey
34:41
Brighton and Hove and West Sussex. Um but at the beginning um my colleagues did a
34:47
lot of work on what was the best place to get the trends from because all we're looking for is trends uh so they
34:52
looked at SEND two, they looked at school census and they did a regional uh version of it as well so they took all the
34:58
the data from the southeast region um on from the school census and SEND two to
35:03
identify the best trends um and ultimately Went with the school census just for East Sussex five-year trend
35:10
so it's probably something if you're going to recreate this is to look at what you think might be the best source of data for it because yes
35:17
you say if you've got a lot of going out of borough you won't know about them in your school census but you will know about them in the
35:23
SEND too. hopefully that answers the question
35:30
>> ALISTAIR: Here's another fun question for us from Ben there. Um other than our first first project um
35:37
do schools being enclosed have an effect on demand at the front door? 
>> SARA: um yes they do um and i can actually
35:44
show you a relevant chart um apologies let me see if i can
35:51
share my screen again is that sharing 
>> ALISTAIR: Yep that's on yeah
35:57
>> Sara: Wonderful. Um so you will see that even from our historical data
36:03
um September tends to, September and October tend to have a pickup in demand in terms
36:10
of the contacts that come to the children's single point of access compared to the summer months um this is
36:18
the line that we would have expected had there been no COVID19 so basically this would have been the
36:24
baseline in terms of the demands that we would expected coming from schools actually our projection was much higher
36:31
and the actual matched our projection now we a big peak coming in September and October
36:37
um which was actually much bigger than what we experienced the previous year um this was also due to the fact that
36:44
obviously schools had been closed for quite a while um during the uh period of the lockdown
36:50
and had not had an opportunity to refer children where there was a concern, plus concerns
36:56
increased in general and um basically when schools reopened
37:02
that meant that there was a lot of pent up demand that was unloaded onto our children single point of access
37:07
>> ALISTAIR: Absolutely. It brings up another point which i had meant to cover in
37:13
in my presentation but didn't. About not being blind when you're modelling to the factors that you're not modelling
37:19
and our first um front door demand modelling tool for lockdown
37:24
ignored the fact that there were seasonal variations in referral volumes for local authorities and so a lot of local
37:31
authorities will see much higher volumes of referrals to social care in October, November time
37:37
than they will in the summer or the kind of the late winter, February kind of time
37:43
and once we applied a kind of historical model to say what's your normal seasonal variation
37:49
we found actually the significance of that in some local authorities was much greater than the change we were
37:54
modelling in volumes because of lockdown and that was interesting and kind of raised that point that you can't you
38:01
can't afford not to be aware of the wider context whether it's something like that and in terms of seasonality
38:07
where it's the kind of big macro-economic changes that are going on in um in the wider world or also whether the
38:15
strategic change is happening locally in a decision someone makes to manage demand in a certain way which i
38:20
think someone asked a question or made a point about in the conversation as well which I’ve
38:26
now lost 
>> SARA: Yeah the way we accounted for that phenomenon was effectively
38:32
definitely looking at seasonality the way it had been observed in previous years because as you say there is a
38:38
strong seasonal component in terms of the opening or closing of schools at the best of times um but also we
38:43
looked at basically, what we termed at the time, as missing contacts so effectively we looked at what our
38:50
normal seasonality would have been in the lockdown months um and then we uh projected the demand
38:56
that we would have normally like we saw we projected the actual demand that we had received
39:02
and we realized that we had had far fewer contacts than we normally would have expected
39:07
so um that was obviously uh partially due to the closure of schools and um the way we took account of that
39:14
in the modelling was effectively using those uh contacts as uh contacts
39:19
that needed to be basically made up for in a sense uh later in the year um and therefore they went to add to the
39:27
projection and that we would have otherwise created for convict 19.
39:34
>> ALISTAIR: I’m just working through the chat channel looking for questions we've missed um but feel free to chip in um
39:41
by chat or by voice anyone who'd like to do so.
>> SPEAKER 1: Alistair there's a question in the chat from
39:49
Neil [unintelligible surname] 
>> ALISTAIR: what calculations are you using for
39:55
forecasting population and how do you factor in population migration etc?
40:01
>> ALASTAIR LEE: Um and the short answer is pop group and i was typing that into the um into the chat again so yes we just
40:08
take the data that's provided by ONS um however LGA manages pop group now
40:15
um we have a team in East Sussex that looks at the population figures as well so we our colleagues in Corporate
40:21
uh deal with that and we get the data from them um so it in a way it can be seen like quite
40:26
a blunt tool you know because it and it’s basic maths that goes up uh but it's the element
40:31
of then presenting that for the challenge with the workers that is the most valuable part because it's it A it gets closer to
40:39
to reality but also it gets them engaged in understanding um and the workshop one of the first
40:45
workshops i was in was really great where the different parts of the team started uh challenging each other about you know
40:50
do we think MLD is going up or is it SEMH and actually because there's a relationship between the two um so
40:56
really interesting and the head of service uh Nathan Kane has been very keen they continue every year so we've
41:02
gone from a day long saying two now it's a half day thing and actually the most recent one was done by teams um because people are used
41:09
to the process but very valuable in terms of not only setting the forecast but getting to understand between the teams what the challenges
41:16
are for the service so it's more than just a forecasting tool 
>> ALISTAIR: one of the interesting things that came
41:21
out of the data to insight demand modelling workshops has been about just how many
41:27
different things there are that you could fold into your model and there are so many factors there and at a certain point you have to kind
41:33
of you do all this open-minded stuff about throwing everything on the paper and figuring out there's all these different things
41:39
at some point you have to close your mind and figure out these are the things that we're going to start by modelling and
41:44
then see where that takes us and if it works and kind of this idea of not letting the kind of the
41:51
limitation of what you're doing stop you using it if it turns out it's
41:56
useful and it does something that works.  it takes us around to um sorry
42:03
Alastair do you want to jump in there
>> ALASTAIR LEE: i was gonna um leap in and my um experience from the send forecasting model um which was
42:11
predates data insight i said was very much it was a leap of faith for the managers to say you know we can do this you know
42:17
so it's basically shut up Alastair and go and do it you know you keep bashing on about the fact that in principle you can do this so
42:22
go and do it and show us what you produce so there is a relationship thing between myself and the service managers about
42:29
them being okay with me producing stuff for them like as you know in our work you can
42:35
produce the chart as much as you like but if nobody reads it or nobody's got any interest and it's pointless so it's
42:40
how you then work through that i think it might have taken quite well it did take quite a while for me to get the service managers okay for me to do
42:47
something in principle as a test which they then saw the value of uh and so it is 
42:53
that's part of the process as you know you can't just build a dashboard and everybody loves it you have to do a bit of work around
42:59
around the edges to get the uptake um and uh one thing i haven't said and it would
43:05
be unfair to miss it out is that we the service managers in the second year commissioned Mastodon C to do our
43:11
do send forecasting using their model uh and they've continued to run both our model and the Mastodon C alongside each other
43:18
for about three or four years um and eventually concluded that they produce very similar information
43:23
and have kept ours and Mastodon C um uh are obviously developed but we're not
43:29
using them anymore but it was really useful to have those two running along side by side so the fact that you know in even this
43:36
conversation we've got almost three versions of looking at uh demanding children's services uh for children looked after um it's
43:43
great you know we have to have the different models and test them out and learn from each other and stuff like that so yeah really good to
43:48
have the different bits.
>> ALISTAIR: It reminds me of a great conversation i had this year with
43:54
someone i won't name about how they could do quite wonderful and remarkable um
44:01
things with maths to do their forecasting but actually the thing they ended up taking to their
44:06
kind of key stakeholder the person who paid the bills was usually a historical line with a
44:12
trend extended because that was the only thing they could get the person who paid the bills to
44:17
to kind of take seriously and worry about and the question i wanted to ask really
44:24
which Alastair you touched on in your presentation I’d be interested to hear from um Indy or Sara if they've got different
44:30
perspectives or similar ones about that conversation with leaders and what matters when you're having that
44:36
conversation what the you know the tricky bits are the key factors that go into the you know the recipe of making something
44:42
that that you can present to a kind of a director level decision maker and say this is what we're going to do because
44:48
my modelling says this is the case. Um how does that conversation work or not work as may be the case.
44:56
>> SARA: In my experience the reason why that conversation worked and it was not totally was because people were involved
45:03
in the creation of the model from the very beginning um so as they had
45:08
helped drive the assumptions and the work that was required for us to produce the model
45:14
they were then more likely to give us the buy-in and the sort of validated the findings and said
45:20
yeah okay so we are going to be driven by what we have participated in creating um and that
45:27
means that what the model is telling us is that we need to do x and z um and that's
45:32
exactly what we are going to do.
>> INDERJIT: i agree at least i think it's i think if
45:38
you've got if you've got that buy-IN at the start and it's part of you know the work forms part of a
45:43
like you know a council or a children's or adults-led program that it kind of it's it's attached
45:49
to something so it's kind of like you know it's key it's there it's visible there's a deliverable
45:55
you come out with a deliverable and actually everyone gets it because actually that's what that's what they're expecting i guess i
46:00
think if you just land it cold and say like here's my model now go and do this
46:06
i think you'll be uh we'll be met with quite a lot of resistance to it.
>> ALISTAIR: Yeah it makes sense.
>> ALASTAIR LEE: Yeah i agree with
46:12
both of those and what you said AlIstair um earlier was um here's a model which explains why we're
46:18
going to make these decisions it's more about here's a model that's going to help you make your decisions you know it's only part of what they're
46:24
going to do there's going to be a lot of discussion in your team and management teams and yeah with your heads of service about the various possibilities but if you've got
46:30
a tool that can help visualize it and maybe tweak some parameters uh very basic uh then that is it's a tool
46:38
that helps it's not all gives answers.
46:44
>> ALISTAIR: Um. There's a bit right up to the top of the chat i missed a question from um Joanne about has anyone done any
46:51
modelling about how needs change over time? This is where i'd like to go with modelling but currently it's just a dream um
46:57
actually there's a team in GMCA that i've spoken to who are doing a kind of they've done a prototype
47:03
system pressures thing and it looks very much like a visualization of conversations i've had with other
47:09
councils about managing the whole flow of cases through their system and seeing that when
47:16
sometimes they might see their child protection population has gone way out of control and then it started
47:21
to come down but as it came down the LAC (Looking after children) population started to rise again and they're kind of building this map
47:27
for their local authorities in Greater Manchester that um that shows you the kind of different levels of service like contact
47:33
referrals assessments, CP, LAC and what have you and showing you over time
47:39
how um how high the volumes are in each of those systems alongside each other um just to show
47:45
that kind of level of pressure and i think that's quite interesting in that area the other the other thing that um
47:51
that's Sara work i think really what excites me about Sara’s work is that potential to say
47:57
well we can kind of see this movement through the through the system through this flow can we use that to say well last
48:03
time we had this many referrals it ended up in that many LAC later or something like that but i think. that's not something that
48:08
i've seen working elsewhere. Joanne you wanted to chip in?
>> Joanne:  Um i was  more thinking about um
48:17
SEO needs so um obviously you get to a point well there's a point along the
48:22
system where you i guess there will be a proportion of children who
48:27
are um SLCN until they're a certain age and then they get a diagnosis and they
48:32
move into the ASD category and it's about understanding what proportion of children that happens for
48:38
and also at roughly what age it might happen and just so you can look at what demand there will be
48:44
in at different ages that's more where i was going 
>> ALISTAIR: Okay can we can we pass it across to
48:49
Alastair who's got his hand up and then come back to Indy um afterwards he'll be talking about something we're talking
48:54
about before possibly 
>> ALASTAIR LEE: Um yes uh just on that one it is very interesting you can see it in historical
49:00
data and equally you can see fashions as well so um as something becomes uh the thing
49:05
that people are being diagnosed with you can see those going up and other things going down there is
49:11
also yes as you say at key points so between the key stages you'll see changes in a diagnosis or a primary
49:19
need um and you'll pick it in pick it up in your data as you do it um that again forms part of the
49:25
discussion on the slide i've got a we have a whopping great slide pack for our presentations
49:30
and in that there is a box that gives the justification for the trend and that's where the frontline staff say
49:36
well we know that you know at year 7 a lot of X happens and that you know
49:42
we're going from prime uh nursery to primary a lot of Y happens and that's why that affects the trend so that a lot
49:47
of that is coming in through practice um that yeah.  Also the reason reason put my hand up is that earlier on um
49:53
Ben Watkins asked about how much of the shift of demand is due to changing threshold rather than lock down
49:58
um i say i would say yes some of it is to do with threshold and that's just understanding the system i
50:04
wouldn't think really. Hopefully that helps on both questions.
50:11
>> ALISTAIR: Yeah thanks Alastair. Indy you stuck your hand up and i just ignored it i'm very sorry. What were you gonna say.
50:16
>> Indy: No that's cool it's just quickly on um i've got two quick points just quickly
50:22
on Alastair’s bit around the growing needs so we did some work in Dudley on SEND so we found ASD,
50:28
SCMH and speech and language as the kind of the primary grow growth areas in Dudley
50:34
as well so it's consistent what we're finding as well as if that's any that's any help um
50:42
and then just quickly i think Ben asked a question quite a while ago and just wanna.
50:48
Hang on a second it was uh what specification can you give of opportunities there are in children's
50:54
commissioning to adults. Ben was that was that linked to what i said Ben or was that
51:00
something else? i've got an answer to it if it is
51:07
>> ALISTAIR: So should we pretend that it was um in absence of a reply?
>> INDERJIT: okay assuming it was then so um
51:14
so i'll give you a quick example of something in adults where you know it's the principles of
51:19
commissioning really so you've got what we've done in Dudley we've done like provide lead models
51:24
for older people so and it's like a model that incentivizes reduced care so older
51:32
people um going to hospital coming out with hospital rehab and went to hospital discharge
51:37
now the whole model was like predicated on like pay you know pay and enhance rate at the
51:43
start with them with the providers so you select a few providers pay the enhanced rate
51:49
skill them to um to to deliver the real money service
51:55
wrap around the support around that that person so OT’s, physios
52:01
technology etc etc with the aim of reducing that person's ongoing care need
52:06
so it's it's like pump priming at the start with the provider that gets it and understands it and wants to do it
52:13
so that's i know it's older people i know it's a totally different demographic but actually the principle of the of the commissioning model shouldn't be
52:20
any different because you know you're talking to providers who should be in
52:25
in kind of like you know in the industry to make a difference really if you can kind of tap into that bit of
52:31
their work get the right ethos providers but also tap into their financial benefits of it as well because everyone
52:38
has to win in this. That's the bit i think around you know there's different ways and adults that that commissioning works and you can you
52:44
can flip it to children's if there's a desire to really
52:51
>> ALISTAIR: Thanks Indy cool. Um can anyone help me out finding any questions i may have
52:58
missed or do we think we're done?
53:07
Logan asked what is the underlying model being used for the forecast? Um but Logan you’ve got your hand up do want to chip in?
53:13
>> LOGAN: yeah sure um i wanted to get a bit more clarity on
53:19
what the underlying model was for the contacts data. Like i understand it's weighted but what the underlying model if you're using a
53:26
regression model or exponential smoothing or Cerema i just wanted to know what the
53:32
underlying engine.
>> ALISTAIR: Was that from Sara’s work or from the data to insight?
53:37
>> LOGAN: i believe so Sara yes.
>> SARA: Right so um in terms of basically uh
53:44
what we have created on the um well basically the engine running
53:50
um at the bottom is a very simple um baseline as it were that would have
53:56
followed the um trend for uh different uh indicators according to the
54:03
data in our possession from the last three years so that was basically the baseline um and then um based
54:11
on the baseline data we modified the baseline data based on the behaviour of
54:16
the different indicators um in lockdown months and um
54:22
we then created basically a return to baseline model um that took the indicator from
54:29
baseline to lockdown period according to the behaviour that we had um experienced in the first lockdown and
54:36
then um from the lockdown period back into the baseline over a period of time when
54:43
we um um basically saw the effect of the lockdown uh waning um and
54:50
obviously the process was repeated for the different indicators according to their
54:55
uh to the different behaviors during lockdown but also according to the impact of the indicators that they were
55:01
based on. does that make sense?
>> LOGAN: Uh mostly um so it's more like a weighted regression
55:08
so you did a you said it was a simple 
>> SARA: Pretty much
>> LOGAN: Okay thank you very much
55:13
>> ALISTAIR: Thanks. i think that is probably a good point for us to sign off before people need to
55:20
leave anyway at three o'clock so i think all that remains is to say thank you very much for coming um
55:27
thanks for useful questions and interesting answers from people hopefully some of that was useful
55:33
um if you would like to catch up with anyone after the session by email or what have you then you'll
55:40
have some contact details somewhere find a way to get in touch and we will find a way to put you in touch with the person you want to speak to. i'm
55:46
usually quite an easy person to find so by all means come by me if you'd like to um
55:52
the data to insight project we're running um development openly in our slack channel for local authority analysts
55:57
so if you'd like to keep tabs on that as it develops then by all means ask me for an invitation to that channel and i will pass it on.
56:04
um yeah sure we can find a way to circulate the slides as well that'll be fine. Um thanks very much for coming thank you
56:11
Alistair, Indy Sara for presenting and it's been really useful for me as well as for participants hopefully thank you
56:20
and thank you Sam for hosting us.
56:33

OFFICIAL