I have worked out all the implicit conversions, but I still see mentions of it in the plan. I have attached the plan, and any recommendation will help.
```
select cardholder_index, sum(value) as [RxCost]
into #rxCosts
from RiskPredictionStatistics with (nolock) where model_name = 'prescription_cost_12_months'
and model_set_name = 'rx_updated' and run_id in (select value from #runIds)
and exists (select 1 from StringContainsHelper with (nolock) where IntValue = cardholder_index and ReferenceId = @stringContainsHelperRefId)
group by cardholder_index
```
https://www.brentozar.com/pastetheplan/?id=SyHl9xFeS
It looks like when you populate the `#runIds` table -- and I'm just taking a wild guess here -- you're using a string splitter function that outputs the values as `NVARCHAR(MAX)`.
[![NUTS][1]][1]
You could try converting the values there to get rid of the implicit conversion warnings.
Another possible improvement would be to alter the `NonClustereIndex-Cardholder` index on RiskProductionStatistics to have `model_set_name` as a key column, and `model_name, run_id, value` as included columns. This would address the Key Lookup.
[![NUTS][2]][2]
[![NUTS][3]][3]
You may also want to check the datatype of `model_name`. It appears in a Filter operator, and my fear is that it's a MAX datatype, which may prevent the [predicate from being pushed down][4].
[![NUTS][5]][5]
Since this is an estimated plan, and you haven't included any metrics about the query, it's hard to say how much improvement these changes will have.
[1]: https://i.stack.imgur.com/8xt66.png
[2]: https://i.stack.imgur.com/9gK1n.png
[3]: https://i.stack.imgur.com/9TDNx.png
[4]: https://www.brentozar.com/archive/2016/10/max-data-types/
[5]: https://i.stack.imgur.com/bP1ZA.png
