Finance is the force that keeps the other five honest.
It accounts for whether any of them are working, and it handles the operational machinery of the business: budgeting, forecasting, reporting and compliance. It generally generates no income directly, barring the returns available through astute capital management and reinvestment. What it determines is whether anyone can tell where the revenue came from, what it cost to earn, and whether the decisions made last year were correct.
This is the final part of six. Across the series we have worked through Demand, Marketing, Sales, Value Delivery and Back Office Operations. This post covers Finance and then does the thing the series has been building toward: turning six diagnoses into one articulated artificial intelligence (AI) strategy that fits on a single page.
What Finance actually is, and the test it imposes
Finance is where the performance of the other five forces becomes visible. Everything else in the business is an assertion until it appears here.
That gives this force a role in your AI strategy that has nothing to do with automating the accounts payable run. Finance is the accountability mechanism for the entire framework, and it imposes a test worth applying before you commit a dollar:
If your AI strategy in another force cannot be seen in the financial statements within a defined period, you do not have a strategy. You have an experiment.
Experiments are legitimate. Budget them as experiments, expect most to fail, and do not present them to your Board as strategy. The distinction matters because organisations routinely spend strategic money on experimental work and then cannot explain what happened to it.
What AI genuinely changes in Finance
Transaction processing and the close. The same agentic pattern described in Part 5 applies directly: reconciliation, coding, matching, exception handling and much of the month-end close. This is real and unglamorous, and it mostly buys back time from people who should be doing analysis rather than data entry.
Forecasting, which is the interesting one. Most mid-market forecasting is extrapolation with optimism applied. The genuine change is not better arithmetic. It is the ability to forecast from operational leading indicators rather than lagging financial ones, and to run scenarios fast enough that the answer arrives while the decision is still open.
Notice that this only works if the other five forces are instrumented. Your forecast improves when it can draw on demand signal from Part 1, pipeline honesty from Part 3 and delivery variance from Part 4. Finance cannot manufacture that information. It can only report what the business bothered to capture.
Working capital. Debtor behaviour, payment timing, inventory positions and cash conversion are all pattern problems with real money attached. For a business pursuing the working capital route described in Part 5, this is where the release becomes visible and bankable.
Controls and anomaly detection. Sampling was always a compromise forced by cost. Testing the full population rather than a sample is now frequently affordable, which changes what assurance means in practice.
What AI does not change here
It does not make a capital allocation decision. Where to concentrate effort remains a judgement about the future, made by people who carry the consequences.
It does not validate a strategy that has no measures attached. If nobody defined what success looks like, no amount of reporting will reveal whether it happened.
And it will confidently produce precise-looking output from poor inputs. A forecast built on data nobody has reconciled is not improved by being generated faster. It is simply wrong sooner and with more decimal places.
The bench strength question
No organisation wins its market because of excellent financial reporting. But the quality of your financial information sets a ceiling on the quality of every decision made in the other five forces.
Test it. How many working days does your close take? Can you see gross margin by job, by customer and by service line without a special exercise? Is your forecast accurate enough that you actually act on it, or is it a document produced for the bank? And can you tell, today, which of your customers are profitable?
If those questions are uncomfortable, that is not a reason to avoid an AI strategy. It is a reason to know that your first move may be a data and process one, and that pretending otherwise wastes a year.
Bringing the six forces together
You now have six diagnoses. The work is turning them into one decision.
Step one: score each force honestly. For each of Demand, Marketing, Sales, Value Delivery, Back Office Operations and Finance, answer two questions. Is this a genuine bench strength, or a weakness? And is there a significant obstacle or a significant opportunity present, or merely room for improvement?
Most organisations find one or two forces where something significant is genuinely in the way, and four where the honest answer is that things could be tidier. That asymmetry is the useful output.
Step two: choose one. This is the hardest discipline in the entire framework and the one most organisations refuse.
A plan that addresses all six forces is not a strategy. It is a budget with ambition attached. Strategy is concentration of effort against a diagnosed obstacle, and concentration means deliberately not doing things that are individually worthwhile. If your AI plan has nine workstreams and no trade-offs, you have not chosen anything.
Three criteria decide it:
Significance. Is the obstacle or opportunity large enough that overcoming it changes your competitive position, rather than improving a metric?
Defensibility. Does the advantage rest on something proprietary, such as your own data, your delivery history or your accumulated operational judgement? If it rests on access to a model or a tool, your competitor can match it next quarter.
Achievability. Can you do this with the people, data and capital you actually have? An unachievable strategy is not ambitious, it is decorative.
Step three: write it down. One page. If it takes more, it is not yet a strategy.
The one page, worked
Here is what a completed articulation looks like. The example is a composite mid-market specialist manufacturer, illustrative rather than real, but the structure is the point and it transfers to any sector.
The diagnosis
Our margin varies by more than twelve percentage points across jobs that look similar at quotation. We cannot predict which jobs will erode until they are substantially complete. The estimating judgement that would prevent this sits with three people, is not documented, and two of them retire within four years.
The strategic objective
Within twenty-four months, margin variance across comparable jobs is reduced to under five percentage points, and estimating accuracy is no longer dependent on any individual.
Guiding policies
- We will capture estimating reasoning as part of the quoting process, not as a separate documentation exercise.
- We will simplify each process before automating any part of it.
- We will not quote work outside our defined delivery envelope, regardless of contract value.
- We will not invest in AI in Marketing or Sales until this objective is met.
Accountable owner
The Chief Operating Officer. Not the technology function, and not a project manager.
Measures
Activity: percentage of quotes with structured estimating rationale captured. Percentage of completed jobs with variance analysis fed back to estimating.
Outcome: margin variance across comparable jobs. Estimating accuracy at quotation versus final. Time for a new estimator to reach acceptable accuracy.
Defensibility test
The advantage rests on our own historical job and estimating data, which no competitor can acquire, and on encoded judgement from people they cannot hire. Replication would require them to live through the same decade of work. This passes.
Achievability check
Ten years of job costing data exists and is reconciled. The three estimators are willing participants. No new capital is required beyond software and a part-time analyst for eighteen months. This is achievable.
That is a strategy. It has a diagnosed obstacle, a concentrated response, explicit trade-offs including one force it deliberately starves, a named human owner, measures that separate activity from outcome, and a defensibility position that survives scrutiny.
Notice what is absent. No list of tools. No mention of a specific vendor or model. No workstream for each department. The technology decisions follow from this page rather than preceding it, which is the correct order and almost never the observed one.
Five ways this goes wrong
It addresses every force. The most common failure. Nothing is concentrated, nothing is traded off, and the programme becomes a portfolio of small improvements that never changes the competitive position.
It has no guiding policies. If your strategy does not tell people what to decline, it will not survive its first conflict with a large opportunity.
The owner is the technology function. Technology can deliver capability. It cannot own a business outcome in a force it does not run. When the Chief Information Officer owns the AI strategy, the organisation has quietly decided this is a technology programme.
The measures are all activity. Adoption rates, licences deployed, hours saved. All of these can improve while the business is unchanged. At least one measure must be an outcome the Board would recognise.
It is not achievable. Written for a business with better data, more capable people or more capital than the one that exists. This is the failure that looks most impressive in the boardroom and dies most quietly.
Where this leaves you
Six posts ago the argument was that most organisations do not have an AI strategy, they have an AI shopping list, and that the gap between them explains most of the disappointing returns in this category.
The framework is the correction. Work through Demand, Marketing, Sales, Value Delivery, Back Office Operations and Finance. Find where the obstacle is genuinely significant. Test whether overcoming it would rest on something a competitor cannot copy. Then concentrate, write one page, and decline the rest.
The single most useful sentence in the whole series is worth repeating. Everyone has access to the same models. Nobody else has your demand signal, your delivery history, your customer relationships or your operational judgement. Any AI strategy whose advantage rests on the model is not defensible, and any strategy built on what only you possess is very hard to attack.
Unisphere works at the convergence of technology, strategy and AI, with the business knowledge to connect them. We have run strategic business and technology planning for organisations across a range of sectors and geographies, and the Six Forces of Business is the framework we use to do it. Our AI and digital transformation practice begins with a readiness assessment and a roadmap rather than a tool selection.
If you have worked through all six forces and want a second opinion on which one to concentrate on, that is a conversation worth having. We should talk.