Someone has sent you this
and wants a decision.
Here is the arithmetic, what else is on your shortlist, how you will know at day 90 whether it worked, and how you get out. In that order, because that is the order you will ask.
AI strategy work in enterprise grocery at nine-figure revenue: change management, orchestration across model providers, and the reporting layer, running against live POS, ERP and inventory with close criteria agreed in writing. Reference calls are arranged while you are still evaluating.
The 200 basis points below. Worked from four signals for a mid-market multi-store estate. Nobody has booked a dollar of it yet. We publish it because the components are checkable, not because it is a result.
We keep these in separate boxes on every page of this site. When the pilot closes we will publish the measured number here, whichever way it went. Pilot status →
Where 200 basis points would come from.
Four signals, none of which require hiring an analyst. Each is a number your own finance team can check against your own history before you believe ours.
The same arithmetic in dollars.
Find your revenue line. The right-hand column is what the modelled upside would have to be worth for the fee to be a rounding error, which is the only ratio a board needs.
| Revenue in scope | Roughly | Setup package | Ward, year one | Modelled at 200 bps | Ratio |
|---|---|---|---|---|---|
| $100M | 50 to 150 stores | Managed | $22K | $2.0M | 91× |
| $300M | 150 to 800 stores | Program | $60K | $6.0M | 100× |
| $750M | 800 to 3,000 stores | Program | $110K | $15.0M | 136× |
The ratio column is modelled against modelled. It is not a return, it is a statement that the fee is small enough that the decision turns on whether the mechanism works, not on the price. That is what the 90 days are for.
On the Ward column. Deploying is free. Ward bills 5% of the model compute that routes through it and nothing else recurring, so year one is the one-time setup package plus an estimate of that fee at this scale, and year two is the fee alone. Revenue and store count are proxies for the package, not the basis: it is set by how many systems have to talk to each other, how many brands you run, and how custom your schema is. How the bill works →
Four options, including the one where you do nothing.
Every board comparison leaves out the failure column. This one does not, including for us.
What happens, and what it costs.
The same four dates and the same number quoted on every other page of this site and in the evaluation packet.
-
48 hoursFirst insight cards
From a read-only connection. Findings on your own data, not a sandbox and not a slide.
-
Week 2Findings ranked by dollars
Stores, SKUs and vendors ranked by what they cost you, with the cause named and the SQL one click under every number.
-
Week 6First gated write
One playbook, one system of record, blocked on an approver role you named. Everything before this is read-only.
-
Day 90The number, either way
Measured KPI delta against the metric agreed on day one. If it did not move, the pilot ends.
Ward makes money one way: 5% of the model compute that routes through it, metered on your own provider keys and billed monthly in arrears. No license, no seats, no platform fee. Setup is a one-time package if you want a named engineer for training and guidance, and it starts at $0.
- 5% of model spend, the only recurring fee
- Setup from $0, invoiced once
- Cancel any time, 30 days notice
How you will know, at day 90, that we were wrong.
A pilot that cannot fail is a subscription with a nicer name. These are agreed in writing before anything connects.
One KPI, named before kickoff
You pick the metric and the threshold it has to clear. It goes in writing before a single system is connected, so there is no value debate at month four and no moving the goalposts at month three.
A baseline we both signed
The pre-pilot number is fixed from your own history, agreed by your finance team, and frozen. We do not get to choose the comparison window after we see the result.
Published either way
At day 90 the delta is written down and sent to you whether it cleared the threshold or not. If it did not move, the pilot ends. We would rather lose the renewal than argue about whether it worked.
Attribution stated honestly
Where a KPI moved for reasons other than Ward, we say so. A closed case cites the SQL behind every number in it, so your team can check the arithmetic rather than take it.
Getting out is not a migration, because there is nothing to migrate.
The real board question about any platform is whether it becomes a dependency. Here is the whole answer.
| Notice period | 30 days | Month-to-month from day 90. No rollover clause, no auto-renew, no termination fee. |
| Your data | Never moved | Federated query against your warehouse. There is no Ward-side copy to delete, and no residency position to unwind. |
| Your policy | Already yours | Nothing binds until someone on your team approves it, and the approvals are on the record. Export them whenever you want. |
| Your findings | Yours to keep | Every case, its SQL, and its audit trail are exportable and stay exported. You keep the artifacts from a pilot that failed. |
| What you lose | The findings, not the infrastructure | Nothing you built stops working. You stop receiving new cards. |
What boards ask that operator pages never answer.
Neither, and the distinction matters. It is a retail-specific layer that runs on the model vendors you already contracted and the warehouse you already built. It does not need an AI strategy to exist first, and it does not become one. The pilot has a KPI and a close date agreed in writing, so at day 90 you have a number rather than a roadmap.
You spend one tier of annual fee, your IT team spends a few hours in week one and a pull-request review in weeks three to five, and at day 90 the KPI has not moved. You stop, you keep the findings and the policy work, and the data never left your warehouse. That is the floor, and it is knowable in advance, which is more than the do-nothing option offers.
You should not, yet. It is modelled arithmetic from four signals, worked for a mid-market multi-store estate, and nobody has booked a dollar of it. We publish it because the components are checkable, not because it is a result. The measured number arrives at day 90 on your own data, and we publish that too.
Less than with any tool that holds your data. There is nothing to migrate: no copy of the warehouse, no proprietary model to retrain, no second user directory. The policies export on request. You would lose the findings, not the infrastructure.
A person you named, in a role you defined, who approved that specific write. Nothing runs unattended, write scope opens one playbook at a time, and the state before and after streams to your SIEM. The accountability model is the one you already run for code changes.
AI strategy work in enterprise grocery at nine-figure revenue: change management, AI orchestration across model providers, and the reporting layer, running against live POS, ERP and inventory. We do not name operators or publish their numbers without permission, so what we offer instead is a reference call arranged while you are still evaluating, which is worth more than a logo. Separately, the askotter platform Ward is built on runs nine companies in production; those are platform customers rather than Ward customers and we keep the two apart on purpose.
The decision is whether the mechanism works on your data.
Not whether the price is right, and not whether the category is real. Ninety days answers it with a number, and the exit is 30 days from any point.
Find out what your data has been hiding.
Tell us about your operation. We’ll show you the problems Ward catches, and the ones your current tools miss.