AI for Sales Operations in Professional Services
A pursuit review sits on Thursday's calendar. The opportunity is a programme at a mid sized insurer, and the question is whether the firm bids. The pursuit file holds a half completed scorecard, a fee estimate in somebody's private spreadsheet, and a resource plan naming a director committed to another client until the quarter ends.
By Thursday, two of the five people needed in the room are on client site. The decision gets taken on a call, the scorecard is completed afterwards to match it, and the proposal goes out at a blended rate one partner agreed verbally.
That is sales operations in a firm whose product is its people. The pipeline is a set of claims about availability, approvals are partner judgements rather than system rules, and revenue arrives across months of delivery, not on the day the engagement letter is signed.
A proposal is a promise about somebody's diary
What the client buys is a named partner, a director and a team, free in a particular window, with the sector experience the brief asked for. So every opportunity carries an implied booking against people promised elsewhere. The pipeline record holds a value and a close date; the resource plan inside the proposal names individuals; the two sit in different systems, reconciled by whoever remembers both. An agent can read the resource plan out of the proposal, match the named people to the resourcing system, and report where a live pursuit depends on someone already committed. It makes the clash visible before the fee letter leaves the building.
The scorecard that gets completed after the decision
Most firms run a go or no go scorecard asking reasonable things: is the client known to us, is the fee quality acceptable, can we deliver this scope, who is the partner sponsor. In practice it is filled in by an associate the day after the partner decided, so it records the decision rather than informing it.
An agent can prepare it from what the firm already holds: prior engagements with this client and the realisation on them, write offs, whether the firm has delivered this scope before and who ran it. Each answer carries its source, so the partner argues with a populated draft instead of an empty form. A no go becomes defensible, and a go leaves a record you can test against the outcome.
Nobody can take the meeting until the register is clear
Before a pursuit can start, the prospect has to clear conflicts and, in audit and advisory firms, independence. The client conflicts register names clients, matters, adverse parties and restricted entities, and the questions that matter are rarely literal name matches. The prospect is a subsidiary of a group another office acts for. The target in the deal is one you audit.
An agent can resolve the entity, walk the ownership tree, search the conflicts register and the matter database for every related name, and hand over a candidate list with the reason each entry surfaced. Clearance is signed by the person whose licence depends on it.
A pursuit file assembled from people who are all on client site
The bid desk is a coordination job wearing a writing job's clothes. A pursuit needs credentials, biographies cut to the client's sector, a method, a fee schedule, references still permitted to be used, and current insurance annexes. Each piece lives with a different person who is billing a client this week.
What the desk actually spends its day chasing
An agent can run the pursuit file as a tracked set of required artefacts, each with an owner and a state, drafted from the last comparable submission where a draft is possible and chased where it is not. It can tell you the proposed engagement leader's biography is two roles out of date, the reference being cited has passed the window the client agreed to, and the fee schedule no longer matches the resource plan in the appendix. Submissions fail on that last mismatch more often than on the method.
The blended rate somebody agreed on a call
Rate cards multiply quietly: a standard card by grade, sector and geography, a client specific card negotiated at the last framework renewal and never reconciled against it, the blended day rate quoted to win a competitive job, and the fixed fee implying a rate nobody has calculated.
An agent can price the proposal from the resource plan and the card that applies, show the implied discount against standard, name the approval that discount requires, and flag where the quoted rate sits below one the same client already holds elsewhere. Granting it stays with the person who owns the margin, now with the number in front of them.
The partner who will never open the CRM
Partners own the relationships, and most regard pipeline admin as a tax on chargeable time. So the record is stale in a pattern: the opportunities that matter least are the best maintained, because somebody with capacity updated them.
Building the record from what already exists
The useful move is to stop asking for data entry. Calendar entries, email threads, the engagement letter, the proposal and the finance ledger already describe the relationship accurately. An agent can draft the opportunity record from those, propose a stage from the artefacts that exist rather than a self reported judgement, and send the partner a confirmation naming what it inferred. Correcting a draft takes a minute. A form takes an evening nobody has.
When the pipeline and the resource plan describe different firms
The sales forecast speaks in weighted value by month. The resourcing forecast speaks in chargeable hours by person by week. The two collide because the pipeline carries no shape: two large wins landing in the same month may be undeliverable.
An agent can translate each late stage pursuit into the demand it would create, grade by grade and week by week, lay that against committed bookings and known leave, and show where the pipeline as forecast cannot be staffed. It works the other way too, flagging a practice forecasting high utilisation with nothing behind it.
Signature is the start of the number, not the number
A fixed fee engagement signed in one quarter that mobilises the next and runs into the following year contributes almost nothing to the period it was won in. Fee bases differ further: milestone, capped time and materials, retainer, contingent. So the sales number and the recognised revenue number disagree for honest reasons.
An agent can build a phased revenue profile for each won and late stage opportunity from the resource plan and the fee basis in the engagement letter, then flag where the two contradict, such as a fixed fee with a plan that burns through it before the halfway milestone.
Work that arrives because somebody owed somebody a favour
A large share of good work comes through referral and alliance. An accountant sends a client. A law firm recommends you on a transaction. A software vendor brings you into an implementation. Each carries obligations: a reciprocal referral, sometimes a fee, often an expectation about which partner fronts the work. Almost none is recorded, so the firm cannot say which relationships generate revenue.
An agent can read the proposal, the engagement letter and the opening thread, propose the source and the introducing contact, and keep a running picture of what each alliance has produced and what has been sent back the other way.
The extension that never appeared as an opportunity
The second phase gets agreed in a delivery meeting. A change order is signed. A scope widens. None of it reaches the pipeline, because the partner reasonably regards it as delivery rather than selling, so the firm's view of its growth rests on new client work while most revenue arrives unforecast.
An agent can watch the delivery estate for the signals: engagements close to their end date with no follow on recorded, change orders signed with no matching opportunity, a client contact appearing in a new set of meetings. Each becomes a prompt to the relationship partner, not a record created behind their back.
The debrief that happens when you lose, and only then
Win loss debriefs follow painful losses and almost never wins, so the firm learns from its worst days and nothing from its best. The obstacle is effort: somebody has to reconstruct what was promised, the fee at each stage, who met whom and how long the clarification cycle ran.
An agent can assemble that pack on closure: the scorecard as it stood at go, the first fee estimate against the final one, the method statement as submitted. Across a year it can show the patterns a single debrief hides: which criteria predict a win, and which the firm scores generously every time.
Where AI fails here
The decision to pursue is partner judgement with a career behind it. An agent can show the fee quality is poor and the right to win thin, and a partner can still bid for reasons about a relationship over the next decade. That call needs a name against it.
Discount and rate authority is not delegable to software. Neither is conflicts and independence sign off, where the firm's licence rather than its margin is at risk. Anything that commits the firm stays with a human: the engagement letter, the named team, the fee, the deliverable promised.
Committing a specific person to a specific window is a people decision, not an optimisation. No system knows who is burnt out, who was promised this client, or who needs this engagement for their promotion case.
Honest forecasting cannot be manufactured downstream. A cleaner roll up of pursuits recorded at the stage that keeps the review quiet is still wrong, and that fix is cultural.
All of this produces exceptions, and exceptions need an owner with time in their week. If nobody clears the queue, it gets cleared in bulk before a board meeting, which is the practice you were replacing.
Where to start
Start with the go or no go pack. The inputs already exist, the output is a draft a partner reviews rather than a decision taken for them, and the value appears the first time a weak pursuit is declined before anyone spends a weekend on it.
Then replay a closed period. Take the pursuits from the last two quarters, let the agent prepare the pack as it would have at the time, and compare its reading against what happened. Agree the acceptance bar first.
Write down first what the agent may read, what it may write back into the pipeline, and which decisions always carry a human name: the pursuit decision, the discount, the conflicts clearance, the named team, the signature.
If your bid desk is one person rebuilding the same document under deadline, describe the work to Adopt AI Partners at hello@adoptaipartners.com for an honest read on which part an agent can carry.