Contents · 7 min read
01 / 05 · The promise
Each proposal promises a better operating result
In short
Better schedules and lower costs are fair aims. Yet the promise alone tells you little. The field-service company must see how the result will enter daily work.
A field-service company starts each day with uncovered work orders. Its lead can also see late technicians, while missing parts may shift the plan. The company is comparing bids for an AI scheduling pilot. Each bidder promises a better result, but the daily gaps make the operating path matter.
The company is comparing bids for an AI scheduling pilot. Each bidder says its tool will improve the result.
That claim may be sound. A scheduling layer can replan when work runs late. It can suggest the best person for each job. An AI-augmented operations function can also flag moves that need human approval.
The demo may also work as shown. That proves the tool can perform a task under test.
But the company buys an operating result. A good demo does not show who will change the rules. It does not show how staff will act each day. The next question is what work must change.
02 / 05 · Daily work
What must change in the daily workflow?
In short
Ask what people will do in a new way each day. Then ask which management choice allows that change. For the field-service company, the answer starts with dispatch rules and approval rights.
Ask the bidder to fill seven boxes. The first holds the promised result. The next names a choice management must make. Then comes the new act in daily work. The fourth names proof. The fifth names what must be in place. The last two name who runs it and who owns it at exit.
For the field-service company, run this test on scheduling.
What changes each morning?
The dispatcher should receive a proposed schedule. It should reflect skills and travel time. It should also reflect working hours.
Which choices stay with people?
Routine moves can follow agreed rules. Overtime still needs a person. So do moves that affect a customer promise.
What must management approve?
Leaders must set the rules for routine changes. They must also give dispatchers the right to act within them.
This is where many pilots fail. The tool works, but the way people work stays the same. Incentives may still favor the old way. No one may own the new flow. Managers may withhold the right to change it. A technology test can look strong while the business stands still.
The impressive demo is still useful. It shows that one part can work. But the field-service company should reject it if no one changes the daily flow. Once that change is clear, the savings can be tested.
03 / 05 · The proof
What makes the promised savings credible?
In short
A savings claim needs proof and clear needs. The proposal must say which records will show the gain. For the field-service company, it must also name the data and management action needed first.
Suppose a bidder says the pilot will cut dispatch work. The claim may be right. The company still needs to know what will prove it.
For the field-service company, work-order records can show what changed. The proposal should state which records it will use. It should also state who checks the result.
The tool needs sound context too. It needs skill records and travel data. It needs working-time rules. It may need parts data and protected bookings.
If that context is absent, the output becomes a guess. AI given no company context can sound able while missing how the firm works. Company knowledge has to be written down and made available.
The same rule applies to leadership. Managers need enough first-hand skill to test the claim. Otherwise, the project may turn into a hollow task. Leaders who do not grasp the tool cannot judge what it needs.
Ask the bidder what happens when a needed record is weak. Ask who fixes it. Then ask which savings still stand if that work runs late.
Case studies also need the same check. A case library full of demos proves that technology ran. That proof has value. It does not show that profit changed.
For the field-service company, a credible bid ties each gain to a record. It names each need and its owner. If the sums hold, one risk remains: what the company owns.
04 / 05 · The exit
Will you own the working capability at exit?
In short
The work should end with a capability the company owns. A license must not sit between the buyer and the asset it paid to build. For the field-service company, the test is whether it can keep running the new flow after the firm leaves.
A working pilot can still create lock-in. The schedule may run well. The staff may use it each day. Yet the asset may stop working when the consulting deal ends.
For the field-service company, the proposal must state who owns the IP. It must also state which license is needed after exit. The same check applies to any required platform.
Ask a plain question. Can your team run and change the work without the consultant? If the answer depends on a new deal, the exit is weak.
A self-contained step gives the buyer a real choice. The next phase can be earned by the result. It should not be forced by access to the asset. GRAIL states this rule in its own terms: each step stands alone and all IP belongs to the client.
Also check the firm's own use of AI. A firm selling this work should show a working system inside its delivery. A method slide cannot take its place. That gap is one of the red flags to test before hiring.
The field-service company now knows what it must own. That leaves one final choice: keep or reject the whole proposal.
05 / 05 · The decision
Reject any proposal that breaks the chain
In short
Keep proposals that link management action to changed work and clear proof. Keep them only when the client owns the result at exit. For the field-service company, one missing link is enough to demand a rewrite.
The company now has a chain it can judge. One bid may still have the best demo. That is useful evidence, but it covers only the tool.
The chosen proposal needs defined work before the pilot starts. It also needs a clear end. Fixed scope and defined deliverables make that test much easier.
Use this review before you sign:
The CEO states the result
Write the operating gain in one plain sentence. Do not accept a broad aim such as transformation.
The operations lead maps the daily change
Name who acts in a new way. Name when that act occurs.
The management team names its choices
Set the rules for routine moves. Set the line for human approval.
The finance lead checks the proof
Name the records that will show the gain. Name who checks them.
The buyer checks what must be in place
List each needed record or rule. Give each one an owner.
The legal lead checks the exit
Confirm who owns the asset. Confirm the right to keep using and changing it.
For the field-service company, every step leads to the next. A dispatch rule enables a new flow. The flow creates records. Those records show whether the gain is real. The company then keeps what works.
If a bidder cannot fill one step, ask for a new draft. If the gap remains, reject the bid. The source pages below show the grounds for each check.
Before you sign
The management team's next task is to test each proposal against the full chain from management choice to changed daily work, then from credible proof to company ownership at exit. If a bidder's draft leaves gaps, a 30-minute conversation with GRAIL gives the team space to examine the specifics and decide what must be resolved before signing. GRAIL will also say honestly whether it can help.
Choose a 30-minute slot. If we are not the right fit, we say so and point you elsewhere.
Questions CEOs ask about this
What red flags should we look for in an AI consulting proposal?
Reject any AI proposal that promises results without showing the path. The promise is credible only if management choices change daily work. The path must also name the proof and company ownership at exit. You must decide whether to demand a rewrite or walk away.
Which vague deliverables should concern us?
A roadmap with no named choice is vague. So is a pilot with no new daily act or named owner. Ask what will exist at the end and who will use it.
When are promised savings or ROI not credible?
Be wary when the proposal does not name the records that prove the gain. The same applies when savings depend on data or management action that no one owns.
What contract language can create IP or platform lock-in?
Watch for a license between your company and the asset it funded. Ask whether you can keep using and changing the work when the deal ends. Apply the same test to any required platform.
How can we spot a technology project disguised as transformation?
Look for a demo with no change to daily work. If the proposal omits new rules and approval rights, it covers the tool alone. The business may stay as it was.
What does the absence of management commitments tell us?
It means the bidder has not shown how the result will happen. For the field-service company, leaders must set dispatch rules and approval lines. Without those choices, the pilot has no path into daily work.
Founder, GRAIL
Johan has spent fifteen years in management consulting and executive leadership, including as Chief Product Officer of a SaaS company. He founded GRAIL to help European leadership teams work out what AI changes at the core of their business. He is the author of The Execution Revolution and the host of the ThinkRoom podcast.
Related reading
Sources
- How to choose an AI consultancy: the red flags around management involvement, endless deals, licenses and proof of profit. (accessed October 8, 2026)
- What does GRAIL cost?: fixed scope and defined deliverables agreed before work starts. (accessed October 8, 2026)
- What does an AI-augmented operations function look like?: how scheduling work splits between the agent layer and people. (accessed October 8, 2026)
- The Margin That Blinds You: why a strong demo can leave the daily workflow unchanged. (accessed October 8, 2026)
- Engineers Went First. You're Not Far Behind.: why company context must be available to AI. (accessed October 8, 2026)
- The AI Leadership Gap Is Already Ridiculous: why leaders need first-hand skill to judge AI work. (accessed October 8, 2026)
- AI Consulting in Stockholm: self-contained steps with all IP owned by the client. (accessed October 8, 2026)
Next scheduled review: February 2, 2027